How to Make the World Add Up: Ten Rules for Thinking Differently About Numbers
Authors: Tim Harford
Publisher: The Bridge Street Press; Альпина Паблишер
Length: —
Tim Harford's ten rules and one golden rule for reading statistics, visualizations, and algorithmic conclusions critically.
Original
TranslationBrief
The essential idea
Harford treats emotional response as a useful warning signal but not a sufficient basis for judgment. Readers should compare top-down statistics with lived experience, clarify definitions before calculating, step back from vivid single events, and inspect the history and context behind a published number.
The later rules ask who is missing from a sample, what results publication bias has hidden, and whether computers and models expose their data provenance and logic. Official statistics and independent datasets deserve attention, while an attractive chart or dashboard never removes the need to inspect definitions, periods, and denominators.
The golden rule is curiosity: seek several explanations, preserve sources and assumptions, and change your mind when new evidence arrives. Technical leaders can apply this discipline to team metrics, architecture comparisons, AI-supported analysis, and stakeholder claims.
Decision lens
Key takeaways
Notice emotional reactions to numbers, then pause before drawing a conclusion.
Definitions, methodology, time window, and sample boundaries come before calculation.
A vivid anecdote should be checked against a broader trend and comparison baseline.
Missing groups and unpublished results can materially reverse an apparent finding.
Models and AI outputs require visible inputs, sources, assumptions, and limitations.
Good evidence can justify changing a previously held view.
Curiosity is the discipline that connects all the statistical checks.
Workplace experiment
Apply it at work
- 1
For a consequential metric, document what is counted, how it is counted, and why it matters.
- 2
Add a sample-coverage and missing-groups check to analytical reviews.
- 3
Show trends and comparison baselines instead of relying on a single point in time.
- 4
Require source links, assumptions, and verification steps for AI-assisted conclusions.
- 5
Schedule a quarterly review of the team's most important evidence-based beliefs.
Choose one action, define the observable effect, and keep the first test small enough to reverse.
Evidence
Sources and further reading
Additional sources
Channel, aggregator, and commentary links confirm the work; they are not the primary source.